Here is an example of a dynamically generated Makefile.build
-that builds two LON-CAPA files (one of which is tth.so.
+that builds two LON-CAPA files (one of which is tth.so).
all: ../homework/caparesponse/capa.so ../modules/TexConvert/tthperl/tth.so
@@ -145,12 +152,141 @@ alwaysrun:
+All that you have to do to alter the behavior of the installation is +edit a single file (doc/loncapafiles/loncapafiles.html). +Adding, removing, and altering files requires proper attention +to the syntax of file format of course. +
+File Format ++The preceding "make build" documentation +gives an example METAGROUP entry describing one particular file. +All data within loncapafiles.html is specified according +to markup tags. The format and syntax of loncapafiles.html +is currently best described by the HTML documentation code at the beginning of +loncapafiles.html (as well as, by example, seeing how various +information is coded). All in all, the syntax is quite simple. +
+Philosophy and notes (the thing nobody reads) ++Packaging the software from CVS onto a machine file system requires many +things: +
+I looked into, and tried, different ways of accomplishing the above +including automake and recursive make. The automake system seemed quite +complicated (and needlessly so in terms of this project since, by and large, +it works to coordinate many different types of build/compilation parameters +whereas we are more concerned with installation parameters). Recursive make +has significant deficiencies in the sense that not all the information +is kept in one place, and there are significant levels of dependency +between all the things that must be done to keep software packaging +up to date. A particularly convincing article I found when looking into +much of this was + +"Recursive Make Considered Harmful" by Peter Miller. Complicating +matters was, at the time, it was unclear as to what categories +of software files we had, and whether or not the directory structure +of CVS would remain constant. With an ever-developing directory structure +to CVS, I preferred to organize the information on a per-file basis +as opposed to a per-directory basis (although there is a successful +implementation of a standard big Makefile in loncom/Makefile). +Additionally, a standard big Makefile assumes certain "normalcy" to +the directory structure of different potential operating system directories +(RedHat vs. Debian). +
++If you take time to look at loncapafiles.html +(and perhaps run the make HTML command) +you will find that the organizing information according to the markup +syntax in loncapafiles.html is simple. Simple is good. +
++loncom/build/parse.pl is the script (invoked automatically +by the various targets in loncom/build/Makefile) that reads +doc/loncapafiles/loncapafiles.html. parse.pl +is capable of reading and returning different types of information +from loncapafiles.html depending on how parse.pl +is invoked. parse.pl has yet to have introduced new sources +of error, and has been tested in quite a number of ways. As with +any parser however, I remain paranoid. +
++My regrets with the current system is that parse.pl is +slow (can take 1 minute to run) and includes a few tidbits of code, +specific to the make process, that probably should be in +loncom/build/Makefile. Additionally, loncapafiles.html +should have a DTD and all those other good SGML-ish things (and parsing +should be done with a real SGML-derived parser). +
++On the plus side, the parse.pl-loncapafiles.html +combination has been working very efficiently and error-free. +
+The current list of configurable files for the LON-CAPA system is +/etc/httpd/access.conf. +All of these configurable files contain machine-specific information. +For instance, the LON-CAPA system relies on unique host IDs such +as msua3, s1, s2, msul1, and 103a1 (specified as a "PerlSetVar lonHostID" +field within /etc/httpd/access.conf). +Non-configurable files simply do NOT have machine-specific information. +The impact on updating software +
+What this means in terms of software updating is that +